A robust method for computing truth-to-track assignments

Multi-sensor, multi-target tracking is the process of tracking one or more targets given periodic measurement reports of the target locations from one or more sensors. When there are many targets and/or the target density is high, this can be a complex problem. Because of the complexity, it is useful to measure how well the multi-target tracking system is performing. There are many ways to measure the performance but usually no one single measure can encompass the overall performance. Much work has been done on defining useful metrics such as tracking accuracy, track purity, and track continuity. Before any of these metrics can be applied, one must first determine the identity of the target tracks, i.e., the target being tracked by each track. Determining which track corresponds to which target is called the truth-to-track assignment problem. In the past, this assignment has been accomplished using local cost or global cost minimization techniques. While these techniques are useful when using real data, we will show that the assignments it produces can be less than ideal. In the cases where the measurement reports can be identified, such as in computer simulation, we will describe a much more powerful technique, called report-identity based (RIB) assignment that is much more robust and objective. Furthermore, RIB assignment can be applied equally well to both report-to-track and track-to-track tracking systems. The paper will describe how it can be used for all types of tracking systems.